Empirical Results

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

Financial Tsunami (FT). During the period of financial tsunami, people will decrease additional expenditures and increase savings because of uncertain incomes and the possibility of unemployment. Hence, the unnecessary tourism expenditure will be lowered and demands for accommodation and F&B in international tourist hotels will be reduced. Therefore, the financial tsunami is expected to have a positive relationship with input slack and a negative relationship with pure technical efficiency.

The definitions of relevant variables are summarized in Appendix 2B. The descriptive statistics of relevant variables is presented in Table 2.1. On the output side, guest room revenues range from 26 million to 1,479 million NT dollars; F&B revenues range from 5 million to 1,250 million NT dollars; other revenues range from 10,758 to 454 million NT dollars. On the input side, guest rooms range from 50 to 873 rooms; labors range from 53 to 982 employees; F&B expenses range from 3 million to 368 million NT dollars; other expenses range from 10 million to 1,085 million NT dollars. These represent that there are extremely different among individual international tourist hotels on the output and input sides.

The guest room is represented as the quasi-fixed input because only 12 out of 47 international tourist hotels change the quantities of guest rooms during the period of 2003-2009 and most international tourist hotels change within 10 guest rooms (see Appendix 2C). In addition, 58.7% of international tourist hotels have 201 to 400 guest rooms, indicating that over half international tourist hotels is the middle size. The average value of market condition dummy indicates that 14.9% of international tourist hotels are resort hotels. The average value of hotel style dummy represents that 59.3% of international tourist hotels are chain hotels.

2.3.2 Empirical Results

Before evaluating the efficiency of international tourist hotels, this paper examines the problem of data errors (or influential observations), the correlation of input and output variables and the choice of input-output mix. Since the existence of data errors will distort the DEA efficiency- evaluation results, the method proposed by Wilson (1995) is used to detect influential observations. The process is briefly described in Appendix 2D. The result shows that no observations play a relatively important role in determining the efficient frontier, because the value of total effect by removing any observation is not too high (see Appendix 2E). Hence, no observations need to be deleted from the data.

Golany and Roll (1989) considered that input and output variables should follow the assumption of “isotonicity”. It means when an increase in any input variable should not result

‧

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

in a decrease in any output variable. This paper applies the Pearson correlation coefficients to examine the isotonicity relationship between input and output variables. The result indicates that input and output variables are positive relationships at the 1% significant level (see Table 2.2). Hence, the input and output variables conform to the assumption of isotonicity.

Since the results of DEA efficiency-evaluation are sensitive to the input-output mixes, this paper utilizes the Pearson correlation coefficients to perform the stability test. Based on the same input variables, four kinds of input-output mixes are chosen. First, Mix 1 is the original mix and includes 3 output variables: guest room revenue, F&B revenue and other revenue.

Second, Mix 2 includes 2 output variables: guest room revenue and F&B revenue. Third, Mix 3 includes 2 output variables: guest room revenue plus other revenue and F&B revenue.

Finally, Mix 4 includes 2 output variables: guest room revenue and F&B revenue plus other revenue. The results show that the efficiency-evaluation results among four kinds of input-output mixes are positive relationships at the 1% significant level (see Table 2.3).

Hence, the choice of input and output variables in the original mix is appropriate.

The first stage. First, the impact of quasi-fixed input is investigated. The comparison between the efficiency measures estimated by the DEA model without the quasi-fixed and adjusted inputs (Model 1) as well as those estimated by the DEA model with the quasi-fixed input and without adjusted inputs (Model 2) is presented in Table 2.4. Furthermore, this paper uses the Wilcoxon signed rank test to examine whether the efficiency measures estimated by Model 1 and 2 are significantly different or not. Table 2.5 shows that the technical efficiency and pure technical efficiency measures estimated by Model 2 are significantly lower than those estimated by Model 1 at the 1% significant level, implying that the DEA models without the quasi-fixed and adjusted inputs overestimate the technical and pure technical efficiencies of international tourist hotels. The scale efficiency measure estimated by Model 2 is higher than that estimated by Model 1 at the 5% significant level, implying that the DEA model without the quasi-fixed and adjusted inputs underestimates the scale efficiency of international tourist hotels. Hence, the necessity of considering the existence of the quasi-fixed input is justified.

The evaluation results for each efficiency measure estimated by Model 2 are summarized in Table 2.4 and are described in the following paragraph. The mean technical efficiency measure of international tourist hotels is 0.791, implying that international tourist hotels in Taiwan could reduce inputs by 20.9%, on average, and still produce the same level of outputs.

In order to investigate the source of technical inefficiency, technical efficiency can be

‧

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

decomposed into pure technical efficiency and scale efficiency. The mean pure technical efficiency measure is 0.835 while the mean scale efficiency measure is 0.946. These results imply that the technical inefficiency mainly results from wasted resources. In addition, the percentage of hotels operating on the frontier is about 11.2 (37 out of 329) in technical efficiency and 15.5 (51 out of 329) in pure technical efficiency. This result implies that an ample space exists for most international tourist hotels in Taiwan to improve their efficiency.

The second stage. The labor, F&B expense and other expense input slacks yielded in the first stage are used as dependent variables, as well as the degree of market concentration, hotel size, market condition, hotel style, SARS and financial tsunami are used as independent variables in the SFA model to purge effects from exogenous variables and statistical noise in the second stage. Before applying the SFA approach, these values of variance inflationary factor (VIF) are calculated to examine the degree of multicollinearity among independent variables. Since the values of VIF are all below 2.45, the multicollinearity problem among independent variables is not serious.¹¹ Following Fried et al. (2002), the likelihood-ratio test (LR test) is applied to examine the specification of SFA model. The null hypothesis of this test is that the SFA model is equivalent to the traditional model, without managerial inefficiency effect. When there is managerial inefficiency effect, the null hypothesis will be rejected and SFA should be applied. Otherwise, the ordinary least square (OLS) regression should be used (Coelli et al., 1998). The LR test results reject the null hypotheses in all input slack equations at the 1% significant level (see Table 2.6). Hence, the SFA model is adequate to be used in the second stage.

The SFA results are presented in Table 2.6.¹² The degree of market concentration has positive effects on all three input slacks at the 1% significant level, implying that the more competitive pressure can help international tourist hotels to increase their pure technical efficiency. The underlying reason is that when there are more competitors in the market, international tourist hotels are more willing to reduce wasted resources in order to survive (Lovell, 1993). The result also supports the quiet life hypothesis proposed by Hick (1935) that if international tourist hotels have more market power, the manager will pay less attention to improving their efficiency. Two hotel size dummies are positive on all three input slacks at the 1% significant level. Moreover, the bigger hotel size is, the larger coefficient

11 The VIF value is smaller than 5 for each independent variable with no serious correlation with each other; but there exists serious multicollinearity problem if the largest VIF value exceeds 10 (Greene, 2000).

12 In order to obtain better result, SARS is deleted from the other expense input slack equation.

‧

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

will be. The result implies that an expansion in the hotel size may increase the complexity of allocating resources more than decrease the input usage through sharing or joint utilization.

Contrary to the expectation, the resort dummy is negative and significant on all three input slack. Two possible explanations for this outcome are that popular visiting spots help international tourist hotels to attract more visitors, or managers of resort hotels may adopt superior managerial strategies to improve their efficiency (Wang and He, 2006). The effects of hotel style are positive on labor and F&B expense input slacks, but is negative on the other expense input slack, indicating that chain hotels could reduce their other expenses by attracting visitors and benefiting from hotel chains’ managerial experience, but could increase their labors and F&B expenses in order to require standard services and facilities. The SARS has positive effects on labor and F&B expense input slacks at the 1% significant level, indicating that demands for accommodation and F&B in international tourist hotels could be decreased in order to avoid SARS infection during the period of SARS prevalence. The financial tsunami dummy is positive on labor and other expense input slacks at the 1%

significant level, implying that people may reduce the unnecessary tourism expenditure in order to face the uncertainty of the economic environment during the period of financial tsunami.

The contribution of managerial inefficiency is also showed in Table 2.6. The estimated values of parameter γ are all close to 1 in three input slack equations.¹³ It means that the inefficiency is mainly due to the managerial inefficiency. Since the variation in input slacks mostly results from the exogenous variables and managerial inefficiency, the impacts of exogenous variables must be eliminated to avoid misleading the efficiency measures of international tourist hotels. Hence, the necessity of adopting the three-stage DEA is justified.

The third stage. The third stage re-evaluates efficiency measures by using the adjusted variable input data calculated in the second stage. Similarly, the test between efficiency measures estimated by Model 2 and estimated by the DEA model with quasi-fixed and adjusted inputs (Model 3) is presented in Table 2.5 to justify the usage of the three-stage approach. This paper also utilizes the Wilcoxon signed rank test to examine whether the efficiency measures estimated by Model 2 and 3 are significantly different or not. The results show that the technical efficiency and scale efficiency measures estimated by Model 3 are

13 When γ is close to 1, the impact of the managerial inefficiency dominates the statistical noise. Contrarily, when γ is close to 0, the impact of the statistical noise dominates the managerial inefficiency (Coelli et al., 1998).

‧

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

significantly lower than those estimated by Model 2 at the 1% significant level, implying that the conventional DEA models overestimate the technical and scale efficiencies of international tourist hotels. The pure technical efficiency measure estimated by Model 3 is higher than that estimated by Model 2 at the 1% significant level, implying that the conventional DEA model underestimates the pure technical efficiency of international tourist hotels. Hence, the necessity of adjusting inputs is justified.

The evaluation results for each efficiency measure estimated by Model 3 are also summarized in Table 2.4. The mean technical efficiency measure is 0.541, implying that international tourist hotels in Taiwan could reduce inputs by 45.9%, on average, and still produce the same level of outputs. In addition, the mean pure technical efficiency measure is 0.990 while the mean scale efficiency measure is 0.546. These pure technical efficiency measures of international tourist hotels are very close to 1 after discarding the effects of exogenous variables and statistical noise. These results imply that the technical inefficiency mainly originates in the inappropriate production scale. A possible explanation for this outcome is that international tourist hotels may take a long time to reach the appropriate scale since adjusting the number of guest rooms to attain the optimal level may spend many adjustment costs in the short run. Finally, 173 out of 329 observations are purely technically efficient and only 20 are technically efficient, implying that most international tourist hotels in Taiwan have an ample space to improve their technical and scale efficiencies.

Efficiency comparison among international tourist hotels with different types of visitors.

This paper also investigates whether international tourist hotels achieve different efficiencies, when they serve different types of visitors. According to Tsaur (2001), when an international tourist hotel serves group visitors more than 75% in total visitors, the international tourist hotel belongs to TYPE 1 that specializes in receiving group visitors; when an international tourist hotel serves individual visitors more than 75% in total visitors, the international tourist hotel belongs to TYPE 2 that specializes in receiving individual visitors. In addition, this paper adds TYPE 3 that simultaneously receives group and individual visitors, when an international tourist hotel serves group and individual visitor both lower than 75% in total visitors. Table 2.7 shows that the mean technical efficiency measure of TYPE 2 is the highest, and that of TYPE 1 is the lowest. The mean pure technical efficiency measures are all close to 1 among three types. This paper also utilizes the Kruskal-Wallis test (K-W test) to examine whether efficiency measures among different types of visitors are significantly different or not. These results show that technical and scale efficiency measures among different types of

‧

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

visitors are significantly different at the 1% significant level, but the pure technical efficiency measure is not significantly different (see Table 2.7). These imply that the difference of technical efficiency among three types of visitors mainly results from the difference of scale efficiency.

Furthermore, the Wilcoxon rank sum test is used to investigate the multiple comparisons of technical and scale efficiencies among three types (see Table 2.8). These results show that these technical and scale efficiency measures between TYPE 1 and TYPE 2 as well as TYPE 1 and TYPE 3 are significantly different at the 1% significant level, but these technical and scale efficiency measures between TYPE 2 and TYPE 3 are not significantly different. These indicate that if an international tourist hotel mainly receives group visitors, the efficiency will be lower. A possible reason for this outcome is that group visitors book rooms and ask the relative services through travel agencies. However, travel agencies have the better bargaining power and technique to reduce prices or request more services. International tourist hotels which mainly receive group visitors may use more quantities of inputs and still produce the same quantities of outputs. Thus, international tourist hotels which mainly receive group visitors perform lower than others.